[Bug]: Dimension of Ollama embedding is 1536 while the expected dimension is 4096 about quivr HOT 5 OPEN

Hey @binhaoIsUsed, nice to see you back! Diving into another challenge, I see? Let's crack this one together.

The error you're encountering is likely due to a mismatch between the vector dimensions expected by the find_brain_closest_query function and the actual dimensions of the vectors being generated by the Embeddings instance used in CustomSupabaseVectorStore.

The find_brain_closest_query function is trying to execute an RPC call with the query_embedding parameter, which is expected to be a vector of a certain dimension. The error message suggests that the function is expecting vectors of dimension 4096, but the vectors being generated by the Embeddings instance are of dimension 1536.

To resolve this issue, you need to ensure that the vectors generated by the Embeddings instance match the expected dimensions. This could involve:

1. Checking the configuration of the Embeddings instance to ensure it's set to generate vectors of the correct size (4096 in this case).
2. If the Embeddings instance is correct and cannot be changed, you may need to adjust the find_brain_closest_query function to accept vectors of size 1536.
3. Alternatively, if the find_brain_closest_query function cannot be changed, you might need to find a way to convert or adapt the 1536-dimensional vectors to the expected 4096-dimensional format, though this could potentially lead to loss of information or accuracy.

Unfortunately, without more information about the specific implementation of the Embeddings class used in CustomSupabaseVectorStore, it's difficult to provide a more specific solution. If you could provide more details about this, it would be helpful in diagnosing and resolving the issue.

Sources

• backend/vectorstore/supabase.py

from quivr.

Add a brain but getting this error even though the brains_vector table is empty

from quivr.

Yeah sorry about that the seeding of the database makes it so it only accepts OpenAI.

Could you delete all lines in the brain_vectors table on localhost:54323 ? And try again the procedure ?

from quivr.

Would like to fix this bug....

from quivr.

from quivr.